Big data analytics capability and social innovation: the mediating role of knowledge exploration and exploitation

The study addresses growing global sustainability concerns that push organizations toward socially oriented innovation. Social innovation can solve societal problems and enhance competitive advantage, but it poses resource challenges in capital, talent, and knowledge. Organizations increasingly adopt big data analytics (BDA) capabilities to alleviate these pressures. Although BDA’s economic performance impacts are well documented, its effects on social innovation remain underexplored. Drawing on organizational information processing theory and organizational learning theory, the study examines whether and how BDA capabilities influence social innovation and posits that knowledge ambidexterity (knowledge exploration and exploitation) mediates this relationship. The research also investigates the joint configuration effects of exploration and exploitation on social innovation, proposing that consistency between them enhances outcomes while imbalance diminishes them. Using data from 354 Chinese high-tech firms, the study aims to clarify BDA’s impact on social innovation and identify optimal knowledge configurations to drive socially oriented outcomes.

The literature review integrates two primary theoretical lenses and related constructs. Organizational information processing theory (OIPT) posits that firms must align information processing capacity with environmental uncertainty through system design. BDA capabilities function as vertical information systems that collect, analyze, and integrate data for improved decision-making and innovation; knowledge management capabilities operate as horizontal systems facilitating external relationship building and internal integration, both crucial for the ambiguous contexts of social innovation. Organizational learning theory conceptualizes exploration (search, experimentation, discovery) and exploitation (improvement, efficiency, execution) as distinct yet complementary learning activities. Knowledge ambidexterity (exploration and exploitation) underpins firm innovativeness and performance. BDA capabilities are defined as management, technology, and personnel capabilities enabling data-driven insights and competitiveness. Prior work links BDA to various innovation types (green, supply chain, business model, eco- and dual innovation) predominantly via dynamic capabilities and resource-based views, with few empirical studies on social dimensions. Social innovation is defined as a practical process where organizations and stakeholders address societal problems to promote justice, improve living conditions, and create social and economic value. Its antecedents span corporate (e.g., strategic orientation, entrepreneurship), social (institutions, environment), and technical factors (IT enablement). Big data is highlighted as an emerging IT tool for social innovation, though the resource allocation mechanisms for data access and analysis remain unclear. Knowledge ambidexterity literature emphasizes the role of IT capabilities in enabling both exploration and exploitation, which are vital to social innovation via external knowledge sourcing and internal transformation. Hypotheses propose positive effects of BDA capabilities on social innovation (H1a–H1c), on knowledge exploration (H2a–H2c), and on knowledge exploitation (H3a–H3c), with exploration and exploitation positively affecting social innovation (H4a–H4b) and mediating BDA’s effects on social innovation (H5a–H5c; H6a–H6c). Finally, configurations of exploration and exploitation are expected to matter: high–high outperforms low–low (H7a); imbalance in either direction reduces social innovation (H7b).

Design: Quantitative, two-wave survey of Chinese high-tech firms employing OIPT and organizational learning theory to test hypothesized relationships among BDA capabilities, knowledge exploration and exploitation, and social innovation. Respondents: CIOs (Time 1) and CEOs (Time 2). Sampling and data collection: Simple random sampling from a local government enterprise database in Beijing, Zhejiang, Jiangsu, and Guangdong. Criteria: firms concerned with BDA and social issues in past five years; valid CEO/CIO emails. At T1, surveys were emailed to CIOs (basic info, BDA capabilities, knowledge ambidexterity); 463 returned, 442 valid. One year later (T2), CEOs of those 442 firms were surveyed about social innovation; 402 returned, 354 valid matched responses. Measures: 7-point Likert scales. Independent variables: BDA management capability (16 items), BDA technology capability (12 items), BDA personnel capability (16 items) adapted from Akter et al. (2016). Mediators: Knowledge exploration (5 items; Cegarra-Navarro et al., 2011) and knowledge exploitation (5 items; Arias-Pérez et al., 2021). Dependent variable: Social innovation (5 items; Adomako & Tran, 2022). Controls: firm age, size, industry. Instrument development: Two-way translation; expert review by IS and strategy scholars; revisions for content validity. Analytic strategy: SPSS and AMOS. Reliability/validity: Cronbach’s α > 0.7; KMO > 0.7; AVE > 0.5; CR > 0.8. Discriminant validity confirmed via Fornell-Larcker. Common method bias mitigated by time-lagged design, anonymity; Harman’s single-factor accounted for 26.42% variance; one-factor CFA fit inferior to measurement model (RMSEA = 0.039, χ²/df = 1.529, IFI/CFI/TLI > 0.9). Hypothesis testing: SEM (AMOS) for main effects (H1–H4); bootstrapped mediation (SPSS) for H5–H6; polynomial regression and response surface analysis for configuration effects (H7), including consistency (Y = X) and inconsistency (Y = −X) lines.

- Sample: 354 matched CIO–CEO responses from Chinese high-tech firms. - Reliability/validity: All constructs met accepted thresholds (α > 0.7, AVE > 0.5, CR > 0.8, good CFA fit); common method bias not serious. - Direct effects on social innovation (SEM): • BDA management capability → social innovation: β = 0.194, p < 0.01 (H1a supported). • BDA technology capability → social innovation: β = 0.161, p < 0.01 (H1b supported). • BDA personnel capability → social innovation: β = 0.299, p < 0.001 (H1c supported). - Effects on knowledge ambidexterity: • On knowledge exploration: BDAMC β = 0.217 (p < 0.01; H2a), BDATC β = 0.315 (p < 0.001; H2b), BDAPC β = 0.295 (p < 0.001; H2c). • On knowledge exploitation: BDAMC β = 0.194 (p < 0.01; H3a), BDATC β = 0.265 (p < 0.001; H3b), BDAPC β = 0.557 (p < 0.001; H3c). - Knowledge ambidexterity → social innovation: • Knowledge exploration: β = 0.134, p < 0.05 (H4a supported). • Knowledge exploitation: β = 0.252, p < 0.001 (H4b supported). - Mediation (bootstrap; 95% CI excludes 0): • BDAMC → KE → SI: indirect = 0.023 [0.003, 0.052] (H5a). • BDAMC → KX → SI: indirect = 0.036 [0.006, 0.073] (H6a). • BDATC → KE → SI: indirect = 0.039 [0.010, 0.074] (H5b). • BDATC → KX → SI: indirect = 0.063 [0.026, 0.110] (H6b). • BDAPC → KE → SI: indirect = 0.035 [0.007, 0.077] (H5c). • BDAPC → KX → SI: indirect = 0.101 [0.044, 0.171] (H6c). - Polynomial regression/response surface (configuration effects; controls include age, size, industry; F = 18.415, R² = 0.413, ΔR² = 0.022): • Consistency line (Y = X): slope = 0.634, p < 0.001; curvature n.s. → higher social innovation when both exploration and exploitation are high vs. both low (H7a supported). • Inconsistency line (Y = −X): slope = −0.202, p < 0.05; curvature = −0.204, p < 0.001 → greater imbalance in either direction reduces social innovation (H7b supported). • When imbalance exists, social innovation is relatively higher when exploitation > exploration. - Additional note: Education industry control showed a negative coefficient (−0.290, p < 0.05) in the polynomial model.

Findings demonstrate that big data analytics capabilities at managerial, technological, and personnel levels significantly enhance firms’ social innovation. This addresses the research gap by extending BDA outcomes beyond economic performance to socially oriented innovation. The mediating roles of knowledge exploration and exploitation reveal the mechanism: BDA enables access to and processing of internal and external knowledge, which, when explored and exploited, improves the generation and implementation of socially impactful products, services, and models. Configuration analysis shows that maintaining a balanced, high level of both exploration and exploitation yields the strongest social innovation performance, while imbalances erode outcomes—though, when imbalance is unavoidable, emphasizing exploitation over exploration is relatively more beneficial. These results integrate OIPT (BDA as vertical information processing) and organizational learning perspectives (knowledge ambidexterity) to explain how firms can convert data capabilities into social value creation, offering guidance on aligning technology, people, and knowledge processes to address complex social problems.

This study empirically establishes that big data analytics capabilities—management, technology, and personnel—positively influence social innovation in high-tech firms and that knowledge exploration and exploitation mediate these effects. It contributes a theoretically grounded mechanism linking BDA to social outcomes and clarifies that a high–high configuration of exploration and exploitation is optimal for social innovation, whereas imbalance is detrimental. Practically, firms should cultivate data-driven managerial processes, invest in robust analytics technologies, and develop analytics talent while building strong knowledge management practices that balance exploration of external information with exploitation of internal knowledge assets. Future research should validate findings across countries and contexts, examine additional mediators/moderators (e.g., strategic orientation), and complement survey data with objective organizational indicators.

- Context specificity: Data are from Chinese high-tech firms; generalizability to other countries or sectors may be limited. Future cross-country and cross-industry studies are recommended. - Omitted mechanisms: The study examines knowledge exploration/exploitation as mediators, but other variables (e.g., strategic orientation, dynamic capabilities, culture) may also mediate or moderate effects. - Self-report measures: Survey-based data may contain subjectivity; future work should incorporate objective firm data (e.g., archival reports) to improve external validity and reduce bias.